MethaDory: Synthetic epigenomes for open cross-platform episignature classifiers

This record contains the preprocessed DNA methylation array data used to develop and validate MethaDory. MethaDory trains cross-platform episignature classifiers for rare genetic disorders on synthetic epigenomes generated in silico, without using data from affected individuals. The data support the manuscript "Synthetic epigenomes for open cross-platform episignature classifiers" (Ferraro et al.). The data come from publicly available Gene Expression Omnibus (GEO) datasets that were available as of 15 July 2025. We included blood-derived samples profiled on the Illumina Infinium HumanMethylation450, MethylationEPIC v1.0 or MethylationEPIC v2.0 BeadChips. These came from unaffected controls and from individuals with rare disorders that have an established episignature. Processing (R v4.5.1): Per-study quality control removed samples for these reasons: failed Illumina BeadArray metrics sex mismatch contamination genetic duplicates non-blood samples predicted vs. reported age difference >10 years age >80 years PCA outliers Signal intensities were normalized with Noob (ChAMP). For some studies with affected individuals, raw .idat files were not available. Their processed GEO data were used for testing only, not for training.After QC, 5,328 unique control individuals remained and 390 affected individuals. Files: betas_[normalization]_new_targets_[GEO Series ID]: gzipped files within the MethaDory-DNAm-Array.zip folder, contain beta-value matrix (CpG × samples). Illumina CpG probe is listed with column name IlmnID. curated-variants&phenotypes-MethaDory-GEO.xlsx: sample annotation (GEO accession, array type, sex, age group, dataset split, diagnosis where applicable) - [README]: file formats and column descriptions Related resources: Code: https://github.com/f-ferraro/MethaDory Trained classifiers: Zenodo [DOI] Usage note: License and terms of use: The processing, harmonization and annotation in this record are released under CC BY 4.0. The underlying DNA methylation data come from public Gene Expression Omnibus (GEO) submissions. Before reuse, check the terms of each original GEO series and its linked publication, and cite the original studies along with this record.